Is this the same as machine-learning classification?

No. Machine learning can help assign facet values, but faceted classification is an information-organization approach.

Can a facet contain a hierarchy?

Yes. Values may have broader and narrower relationships where those relationships are meaningful.

Must every item have every facet?

Not necessarily. The scheme should define required, optional and inapplicable values explicitly.

AI Glossary / Data engineering and architecture / Beginner

Faceted classification

Definition

Faceted classification organizes information through several distinct characteristics, called facets, that can be combined to describe or retrieve an item. A learning resource might be classified by topic, audience, format and difficulty rather than placed in just one category.

What it means

A facet represents a particular aspect of the material. Topic describes what a resource concerns; format describes whether it is an article, video or another medium. Keeping these aspects separate avoids making a category such as “beginner marketing videos” do several jobs at once.

An item can therefore participate in several meaningful groupings. The same introductory video might belong to the marketing topic, the beginner level and the video format. Users can approach it through whichever characteristic matters to their task.

Faceted classification is the organization of the information. Faceted search is an interface that can use that organization to let people narrow or broaden results. The classification can exist without a particular search interface, and a row of filters does not by itself establish a well-designed classification scheme.

How it works

Designers first identify the important aspects of the collection and define the values allowed within each facet. They document what each value means, how ambiguous cases are handled and whether an item may have multiple values.

Values can form hierarchies. A topic facet might contain business, then marketing, then customer research. Other facets may use a flat list. The relationships within a facet should remain coherent: “video” is a format, not a narrower kind of marketing.

Cataloguers or software assign values to items. Automated suggestions need evaluation because a confident label can still misrepresent the material. Missing metadata should remain distinguishable from a confirmed value such as “not applicable.”

A retrieval interface combines selected values according to explicit rules. A common design uses AND between facets and OR between selected values within one facet, but this is an interface choice that should be made clear. Counts and available options can update as the selection changes.

Three practical examples

A learning library groups resources by topic, learner level and format so that visitors can find introductory videos about automation.
A product catalog separates material, size and intended use instead of creating a separate category for every possible combination.
A research archive describes documents by subject, publication type and period, supporting several routes to the same record.

Think of it this way

Think of describing a meal by its cuisine, main ingredient and serving temperature. These are different aspects of the same meal, so a person can choose a starting point without being forced into one fixed sequence.

The analogy is limited because an information system needs explicit definitions and consistent assignments. Everyday descriptions are often flexible or disputed; useful facets require those ambiguities to be managed.

Worked example

Combining two aspects of a resource collection

An invented library contains 100 resources. Thirty are videos and forty are marked beginner level. The catalog records that twelve resources have both values.

Selecting format = video returns thirty resources. Adding level = beginner with AND returns the twelve in the intersection. The answer cannot be calculated from thirty and forty alone: the overlap must be known from the records.

Suppose the interface then permits video OR audio within the format facet. If eight additional beginner resources are audio only, the result becomes twenty distinct resources. An item classified as both video and audio must still be counted once in the result set.

This example separates classification from retrieval logic. The facet values describe the items; the selection rules determine which combination is returned. It also shows why adding facet counts without accounting for overlap can give the wrong total.

Illustrative case study

A fictional business-learning publisher builds a catalog of AI guides. Its original categories mix audience, subject and medium, producing labels such as “manager videos” beside “data quality.” Editors struggle to decide where a guide belongs.

The publisher introduces separate audience, topic and format facets. It allows more than one topic where justified and writes a short definition for each audience level. An AI assistant proposes assignments, while editors review uncertain cases and sample apparently straightforward ones.

Testing reveals that some visitors interpret “advanced” as technical complexity while others interpret it as seniority. The team separates learner difficulty from job role rather than retaining one ambiguous label. The case illustrates how useful facets depend on shared meaning, not simply on adding more filters.

Common misconceptions and mistakes

Every item needs one exclusive category. Facets allow several aspects, and some facets permit multiple values.
Facets must be statistically independent. They represent distinct characteristics; their values can be correlated in the collection.
A missing label proves the characteristic is absent. Missing metadata may mean the item has not been assessed.
More facets always improve discovery. Unclear or rarely useful facets can make navigation harder.

Business use

Faceted classification helps organizations make growing catalogs understandable without building a category for every combination of properties. It can support product discovery, knowledge management and consistent content planning.

Choose facets from real user tasks and the information that can be maintained reliably. Track whether people can locate relevant items, where they encounter empty results and which labels they misunderstand. A technically correct index can still be difficult to use.

For AI-assisted cataloging, retain the distinction between a proposed assignment and an accepted editorial value. Changes to facet definitions may require reviewing existing assignments, not merely renaming a button.

Check your understanding

A catalog contains thirty videos and forty beginner resources. Can you conclude that twelve items are beginner videos without examining their overlap?

Answer: No. The separate totals do not determine the intersection. The item assignments must establish which resources meet both conditions.

Related terms

Metadata
Taxonomy
Controlled vocabulary
Faceted search
Information retrieval
information architecture
classification scheme

Learn next

Read controlled vocabulary and metadata. Take ten resources, propose three clearly distinct facets, and test whether two people assign their values consistently. Investigate disagreements before expanding the catalog.

Sources

Aida Slavic, Faceted classification: management and use: https://arxiv.org/abs/1705.07047. Author-deposited research on managing faceted schemes; abstract and bibliographic information reviewed.

UC Berkeley, Flamenco Metadata Construction Guide: https://bailando.berkeley.edu/flamenco-metadata-guide.html. Primary project guidance on facet structures and item assignments.

Yee, Swearingen, Li and Hearst, Faceted Metadata for Image Search and Browsing: https://bailando.berkeley.edu/papers/flamenco03.pdf. Primary research on an interface using hierarchical faceted metadata. The library calculations and publisher case are original illustrations.

Previous: Facet theory Next: Faceted search